Snowflake right this moment introduced dynamic mannequin routing inside Cortex AI Gateway, designed to assist enterprises scale back pointless AI spend and enhance what Snowflake calls intelligence effectivity, measuring how successfully firms flip compute, fashions, information, and context into enterprise affect.
The brand new capabilities construct on Cortex AI Gateway, which Snowflake introduced in July 2026 as a unified basis for governing agent connections, intelligently routing requests, and optimizing AI consumption. As firms deploy extra AI apps and brokers into manufacturing, utilizing the identical mannequin for each activity can drive up prices, whereas evaluating and managing a rising mixture of fashions creates extra work for improvement groups. Snowflake is addressing each challenges by making mannequin choice extra computerized and giving prospects entry to a broader vary of open supply and proprietary fashions.
Cortex AI Gateway with dynamic routing can routinely choose the particular mannequin with the optimum stability of high quality and price for the duty at hand. Dynamic mannequin routing can be built-in throughout Snowflake’s flagship AI merchandise, together with Snowflake CoCo and Snowflake CoWork, and is obtainable to third-party AI brokers utilizing Cortex AI Gateway. This new functionality directs lower-complexity or repetitive duties to extra environment friendly fashions, whereas work that requires deeper reasoning is routed to frontier fashions. This helps prospects scale back pointless inference spend, with out having to handle mannequin choice for every request themselves. Snowflake will even be increasing prospects’ entry to main open fashions, together with DeepSeek-V4-Flash 0731 and GLM-5.3, via Snowflake Cortex AI. Along with fashions from suppliers together with Anthropic, OpenAI, Google, SpaceXAI, Meta, and Mistral, this portfolio provides prospects extra freedom to decide on the precise mixture of efficiency and price for every workload.
“Enterprises have gotten far more rigorous in regards to the economics of AI. The query is now not how a lot AI they’re utilizing, however whether or not that AI is translating into significant enterprise worth,” mentioned Sridhar Ramaswamy, CEO, Snowflake. “Attaining intelligence effectivity requires the flexibleness to make use of the perfect mannequin for every activity because the panorama evolves. Snowflake’s position is to soak up that complexity so prospects can deal with outcomes whereas we optimize mannequin alternative beneath.”
Snowflake’s inside testing signifies that utilizing a mixture of open and proprietary fashions for varied duties can ship comparable high quality, whereas materially bettering token effectivity. In a single analysis, brokers utilizing dynamic mannequin routing with Cortex AI Gateway constructed a dbt pipeline with as much as 3x better token effectivity than a frontier-model-only path, whereas sustaining the identical high quality. In a separate check, engineering groups accomplished the identical variety of pull-requests with 25 % better token effectivity.